TECHNICAL COMMENTS
Interannual Variability in Net
Primary Production and
Precipitation
Knapp and Smith (1) suggested that interannual variability in aboveground net primary production (ANPP) is not related to
fluctuations in precipitation, based on analysis of data from 11 Long-Term Ecological
Research sites across North America. This
finding, if applicable to other regions, is
crucial to climate change research, because
it may necessitate revisions of projections
of ecosystem responses to climate change
(2, 3). To examine the relationship between
variability in net primary production (NPP)
and precipitation at a broad scale, a longterm normalized difference vegetation index (NDVI) data set derived from the Advanced Very High Resolution Radiometer
(AVHRR) of the National Oceanic and Atmospheric Administration (NOAA), coupled with a historical climate data set,
should constitute a useful and powerful
data source, because NDVI data are strongly correlated with terrestrial NPP and are
frequently used as NPP predictors (4, 5).
We used an annual mean NDVI data set
over China to quantify temporal NPP variability relative to precipitation variation,
and used coefficient of variation (CV) to
express the magnitude of interannual variability in NDVI and precipitation. We then
calculated CVs of these two variables for
each pixel, with a resolution of 0.1° latitude
by 0.1° longitude, for five biome groups
across China—forest, grassland, desert, alpine vegetation, and cropland (6 )—using
1982 to 1999 NDVI and precipitation data
compiled in China (7 ). We assumed that
interannual variability in NDVI or NPP was
related to temporal variability in precipitation if the correlation between CVs for
NDVI or NPP and precipitation were identified as statistically significant.
The CV value of NDVI for these five
biome groups showed a large spatial variation, with a mean CV of 8.3% for the
forest biome group, 10.4% for grasslands,
24.6% for desert areas, 12.7% for alpine
vegetation, and 9.3 % for cropland. The
largest variation occurred in the desert bi-
ome, followed by herbaceous vegetation
(grasslands and alpine meadows); forests
were the least variable. These results agree
with those of Knapp and Smith (1). However, our statistical analysis also showed a
significant positive correlation between the
CV of NDVI and that of precipitation for
all five biome groups (Fig. 1, A to E). The
coefficient of correlation (r) was 0.43 for
forest, 0.56 for grassland, 0.37 for desert,
0.31 for alpine vegetation, and 0.39 for
cropland, with a strong correlation between
mean CV of NDVI and that of precipitation
for these five biome groups [r ⫽ 0.95, p ⫽
0.012 (Fig. 1F)]. Moreover, the relationship
between CV of NPP estimated based on the
Carnegie-Ames-Stanford Approach (CASA)
model (8, 9) and that of precipitation revealed
trends similar to those implicit in Fig. 1. The
r values were estimated at 0.53 for forest,
0.54 for grassland, 0.48 for desert, 0.37 for
alpine vegetation, and 0.35 for cropland, with
a highly significant correlation between mean
NPP CV and mean precipitation CV for these
five biome groups (r ⫽ 0.97, p ⫽ 0.005).
These results are generally consistent with
those of a previous study (10), but disagree
with the conclusions of Knapp and Smith (1).
Although the data used in the analysis by
Knapp and Smith (1) were from the entirety of
North America and included different terrestrial
biomes, specifically forests, grasslands, and
deserts, their study was limited to 11 sites.
Considering the small sample size and the large
spatial variation of NPP, we suggest that the
conclusions of Knapp and Smith need broader
investigation. Our results, which are based on
remote-sensing approach, suggest that the relationship between interannual variability in NPP
and precipitation across China is the opposite of
the trends observed by Knapp and Smith (1) in
North America.
Jingyun Fang
Shilong Piao
Zhiyao Tang
Department of Urban and
Environmental Sciences and
Key Laboratory for Earth Surface Processes
Ministry of Education
Peking University
Beijing 100871, China
E-mail: [email protected]
Changhui Peng
Ministry of Natural Resources
Ontario Forest Research Institute
1235 Queen Street East
Sault Ste. Marie, Ontario P6A 2E5, Canada
Fig. 1. Relationships between annual precipitation CV and NDVI CV across China for (A) forest, (B)
grassland, (C) desert, (D) alpine vegetation, and (E) cropland; (F) relationship between mean NPP
CV versus mean precipitation CV of all five biome groups. The relationship was significant for all
groups.
www.sciencemag.org SCIENCE VOL 293 7 SEPTEMBER 2001
Wei Ji
Department of Geosciences
University of Missouri, Kansas City
5100 Rockhill Road
Kansas City, MO 64110 –2499, USA
1723a
TECHNICAL COMMENTS
References and Notes
1. A. K. Knapp, M. D. Smith, Science 291, 481 (2001).
2. T. R. Karl, R. W. Knight, N. Plummer, Nature 377, 217
(1995).
3. D. S. Schimel et al., Science 287, 2004 (2000).
4. C. B. Field, J. T. Randerson, C. M. Malmstrom, Remote
Sens. Environ. 51, 74 (1995).
5. S. D. Prince, S. N. Goward, J. Biogeogr. 22, 815
(1995).
6. In (1), biomes were divided into forest, grassland, and
desert. In our work, in addition to these three groups,
the vegetation of the huge highland area of the
Tibet-Qinghai Plateau was handled as a single group,
termed alpine vegetation; cropland was also considered as a separate group, because it constitutes vegetation in a non-natural setting.
7. NDVI data for China were derived from the NOAA/
NASA Pathfinder AVHRR land data set at 8 km spatial
resolution and 10-day intervals from January 1982 to
December 1999. Climatic data (monthly mean temperature and monthly precipitation) were compiled
from the 1949 to 1999 temperature and precipitation
data set of China at 0.1o ⫻ 0.1o resolution, produced
by interpolating data of monthly mean temperature
and monthly precipitation from 682 climatic stations
(11).
8. C. S. Potter et al., Global Biogeochem. Cycles 7, 811
(1993).
9. Solar radiation data used in the CASA model were
derived from 1982 to 1999 Solar Radiation Dataset
of China, produced from 98 solar radiation observation stations across the country.
10. H. N. Le Houerou, R. L. Bingham, W. Skerbek. J. Arid
Environ. 15, 1 (1988).
11. J. Y. Fang, S. L. Piao, J. H. Ke, Z. Y. Tang, Technical
Report, Center for Ecological Research and Education,
Peking University, No. 01-03, 2001.
12. Supported by grant G2000046801 of the State Key
Basic Research and Development Plan and grants
39425003 and 40024101 of the National Natural
Science Foundation of China to J. Y. Fang. We thank
L. Buse for editing an earlier draft.
7 May 2001; accepted 20 July 2001
Response: We appreciate the NDVI and NPP
analyses performed by Fang et al. Additional
exploration of the relationships between climate variability and important ecosystem
processes such as NPP is certainly needed,
1723a
and we agree that the large spatial extent and
sample size available from satellite and climate data sets provides a real opportunity to
robustly test predictions about those relationships. This is one of the recognized strengths
of satellite data sets.
However, we are concerned by a key assumption of their analysis—that NDVI data
can be used to quantify NPP dynamics with
equal accuracy and sensitivity across all biomes. It has been well established that NDVI
can be related to chlorophyll content, leaf
area, and standing crop biomass in most biomes (1, 2) and also NPP in some instances.
Because standing crop biomass and NPP are
positively related across broad spatial scales,
it is common in the remote sensing literature
for these very different ecosystem attributes
to be treated as synonymous. It should be
noted, however, that NDVI-based relationships typically are calibrated with standing
crop biomass data, not NPP. Unfortunately,
NDVI-NPP relationships are not robust under
many conditions. In grazed grasslands, for
example, where standing crop is low but NPP
is high, NDVI can only accurately estimate
standing crop (3, 4). Worldwide, it is likely
that a majority of the grasslands remotely
sensed are grazed.
We are unaware of any studies that have
demonstrated that interannual variability in
NDVI is sufficiently sensitive to detect differences in NPP equally well across the range
of biomes included in the analysis of Fang et
al. Indeed, the sensitivity of NDVI to interannual rainfall variation has been shown to be
low in both very wet and very dry regions of
Southern Africa (5). Thus, we believe that the
conclusions reported by Fang et al. should be
viewed with interest, but also with caution.
The final relationship they present (their figure 1F), which is most relevant to our study
(6), is primarily driven by a single point for
the desert biome, and background soil reflectance in arid regions further complicates
NDVI-NPP relationship in deserts (5). Furthermore, a similar analyses for both North
America and Africa found either no relationship or only a weak relationship between
climate variation and vegetation activity as
determined by NDVI values (7). Although
Fang et al. have used an extensive data set in
their analysis, the strength of our study (6) is
that it was based on direct measurements of
NPP using techniques specifically developed
for each biome. Clearly, the two approaches
are complementary, but the inherent tradeoffs between data quality and spatial extent
must be considered when comparing these
relationships.
Alan K. Knapp
Melinda D. Smith
Division of Biology
Kansas State University
Manhattan, KS 66506, USA
E-mail: [email protected]
References
1. S. N. Goward et al., Remote Sens. Environ. 47, 190
(1994).
2. L. L. Tieszen et al., Ecol. Appl. 7, 59 (1997).
3. M. I. Dyer, C. L. Turner, T. R. Seastedt, Ecol. Appl. 1,
443 (1991).
4. C. L. Turner et al., J. Geophys. Res. 97, 18855 (1992).
5. Y. Richard, I. Poccard, Int. J. Remote Sens. 19, 2907
(1998).
6. A. K. Knapp, M. D. Smith, Science 291, 481 (2001).
7. S. N. Goward, S. D. Prince, J. Biogeogr. 22, 549
(1995).
19 June 2001; accepted 20 July 2001
7 SEPTEMBER 2001 VOL 293 SCIENCE www.sciencemag.org